xref: /linux/kernel/locking/rtmutex.c (revision 764e77d868a5b932c709e20ddb5993f9111a841c)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * RT-Mutexes: simple blocking mutual exclusion locks with PI support
4  *
5  * started by Ingo Molnar and Thomas Gleixner.
6  *
7  *  Copyright (C) 2004-2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
8  *  Copyright (C) 2005-2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com>
9  *  Copyright (C) 2005 Kihon Technologies Inc., Steven Rostedt
10  *  Copyright (C) 2006 Esben Nielsen
11  * Adaptive Spinlocks:
12  *  Copyright (C) 2008 Novell, Inc., Gregory Haskins, Sven Dietrich,
13  *				     and Peter Morreale,
14  * Adaptive Spinlocks simplification:
15  *  Copyright (C) 2008 Red Hat, Inc., Steven Rostedt <srostedt@redhat.com>
16  *
17  *  See Documentation/locking/rt-mutex-design.rst for details.
18  */
19 #include <linux/sched.h>
20 #include <linux/sched/debug.h>
21 #include <linux/sched/deadline.h>
22 #include <linux/sched/signal.h>
23 #include <linux/sched/rt.h>
24 #include <linux/sched/wake_q.h>
25 #include <linux/ww_mutex.h>
26 
27 #include <trace/events/lock.h>
28 
29 #include "rtmutex_common.h"
30 #include "lock_events.h"
31 
32 #ifndef WW_RT
33 # define build_ww_mutex()	(false)
34 # define ww_container_of(rtm)	NULL
35 
36 static inline int __ww_mutex_add_waiter(struct rt_mutex_waiter *waiter,
37 					struct rt_mutex *lock,
38 					struct ww_acquire_ctx *ww_ctx,
39 					struct wake_q_head *wake_q)
40 {
41 	return 0;
42 }
43 
44 static inline void __ww_mutex_check_waiters(struct rt_mutex *lock,
45 					    struct ww_acquire_ctx *ww_ctx,
46 					    struct wake_q_head *wake_q)
47 {
48 }
49 
50 static inline void ww_mutex_lock_acquired(struct ww_mutex *lock,
51 					  struct ww_acquire_ctx *ww_ctx)
52 {
53 }
54 
55 static inline int __ww_mutex_check_kill(struct rt_mutex *lock,
56 					struct rt_mutex_waiter *waiter,
57 					struct ww_acquire_ctx *ww_ctx)
58 {
59 	return 0;
60 }
61 
62 #else
63 # define build_ww_mutex()	(true)
64 # define ww_container_of(rtm)	container_of(rtm, struct ww_mutex, base)
65 # include "ww_mutex.h"
66 #endif
67 
68 /*
69  * lock->owner state tracking:
70  *
71  * lock->owner holds the task_struct pointer of the owner. Bit 0
72  * is used to keep track of the "lock has waiters" state.
73  *
74  * owner	bit0
75  * NULL		0	lock is free (fast acquire possible)
76  * NULL		1	lock is free and has waiters and the top waiter
77  *				is going to take the lock*
78  * taskpointer	0	lock is held (fast release possible)
79  * taskpointer	1	lock is held and has waiters**
80  *
81  * The fast atomic compare exchange based acquire and release is only
82  * possible when bit 0 of lock->owner is 0.
83  *
84  * (*) It also can be a transitional state when grabbing the lock
85  * with ->wait_lock is held. To prevent any fast path cmpxchg to the lock,
86  * we need to set the bit0 before looking at the lock, and the owner may be
87  * NULL in this small time, hence this can be a transitional state.
88  *
89  * (**) There is a small time when bit 0 is set but there are no
90  * waiters. This can happen when grabbing the lock in the slow path.
91  * To prevent a cmpxchg of the owner releasing the lock, we need to
92  * set this bit before looking at the lock.
93  */
94 
95 static __always_inline struct task_struct *
96 rt_mutex_owner_encode(struct rt_mutex_base *lock, struct task_struct *owner)
97 	__must_hold(&lock->wait_lock)
98 {
99 	unsigned long val = (unsigned long)owner;
100 
101 	if (rt_mutex_has_waiters(lock))
102 		val |= RT_MUTEX_HAS_WAITERS;
103 
104 	return (struct task_struct *)val;
105 }
106 
107 static __always_inline void
108 rt_mutex_set_owner(struct rt_mutex_base *lock, struct task_struct *owner)
109 	__must_hold(&lock->wait_lock)
110 {
111 	/*
112 	 * lock->wait_lock is held but explicit acquire semantics are needed
113 	 * for a new lock owner so WRITE_ONCE is insufficient.
114 	 */
115 	xchg_acquire(&lock->owner, rt_mutex_owner_encode(lock, owner));
116 }
117 
118 static __always_inline void rt_mutex_clear_owner(struct rt_mutex_base *lock)
119 	__must_hold(&lock->wait_lock)
120 {
121 	/* lock->wait_lock is held so the unlock provides release semantics. */
122 	WRITE_ONCE(lock->owner, rt_mutex_owner_encode(lock, NULL));
123 }
124 
125 static __always_inline void clear_rt_mutex_waiters(struct rt_mutex_base *lock)
126 	__must_hold(&lock->wait_lock)
127 {
128 	lock->owner = (struct task_struct *)
129 			((unsigned long)lock->owner & ~RT_MUTEX_HAS_WAITERS);
130 }
131 
132 static __always_inline void
133 fixup_rt_mutex_waiters(struct rt_mutex_base *lock, bool acquire_lock)
134 	__must_hold(&lock->wait_lock)
135 {
136 	unsigned long owner, *p = (unsigned long *) &lock->owner;
137 
138 	if (rt_mutex_has_waiters(lock))
139 		return;
140 
141 	/*
142 	 * The rbtree has no waiters enqueued, now make sure that the
143 	 * lock->owner still has the waiters bit set, otherwise the
144 	 * following can happen:
145 	 *
146 	 * CPU 0	CPU 1		CPU2
147 	 * l->owner=T1
148 	 *		rt_mutex_lock(l)
149 	 *		lock(l->lock)
150 	 *		l->owner = T1 | HAS_WAITERS;
151 	 *		enqueue(T2)
152 	 *		boost()
153 	 *		  unlock(l->lock)
154 	 *		block()
155 	 *
156 	 *				rt_mutex_lock(l)
157 	 *				lock(l->lock)
158 	 *				l->owner = T1 | HAS_WAITERS;
159 	 *				enqueue(T3)
160 	 *				boost()
161 	 *				  unlock(l->lock)
162 	 *				block()
163 	 *		signal(->T2)	signal(->T3)
164 	 *		lock(l->lock)
165 	 *		dequeue(T2)
166 	 *		deboost()
167 	 *		  unlock(l->lock)
168 	 *				lock(l->lock)
169 	 *				dequeue(T3)
170 	 *				 ==> wait list is empty
171 	 *				deboost()
172 	 *				 unlock(l->lock)
173 	 *		lock(l->lock)
174 	 *		fixup_rt_mutex_waiters()
175 	 *		  if (wait_list_empty(l) {
176 	 *		    l->owner = owner
177 	 *		    owner = l->owner & ~HAS_WAITERS;
178 	 *		      ==> l->owner = T1
179 	 *		  }
180 	 *				lock(l->lock)
181 	 * rt_mutex_unlock(l)		fixup_rt_mutex_waiters()
182 	 *				  if (wait_list_empty(l) {
183 	 *				    owner = l->owner & ~HAS_WAITERS;
184 	 * cmpxchg(l->owner, T1, NULL)
185 	 *  ===> Success (l->owner = NULL)
186 	 *
187 	 *				    l->owner = owner
188 	 *				      ==> l->owner = T1
189 	 *				  }
190 	 *
191 	 * With the check for the waiter bit in place T3 on CPU2 will not
192 	 * overwrite. All tasks fiddling with the waiters bit are
193 	 * serialized by l->lock, so nothing else can modify the waiters
194 	 * bit. If the bit is set then nothing can change l->owner either
195 	 * so the simple RMW is safe. The cmpxchg() will simply fail if it
196 	 * happens in the middle of the RMW because the waiters bit is
197 	 * still set.
198 	 */
199 	owner = READ_ONCE(*p);
200 	if (owner & RT_MUTEX_HAS_WAITERS) {
201 		/*
202 		 * See rt_mutex_set_owner() and rt_mutex_clear_owner() on
203 		 * why xchg_acquire() is used for updating owner for
204 		 * locking and WRITE_ONCE() for unlocking.
205 		 *
206 		 * WRITE_ONCE() would work for the acquire case too, but
207 		 * in case that the lock acquisition failed it might
208 		 * force other lockers into the slow path unnecessarily.
209 		 */
210 		if (acquire_lock)
211 			xchg_acquire(p, owner & ~RT_MUTEX_HAS_WAITERS);
212 		else
213 			WRITE_ONCE(*p, owner & ~RT_MUTEX_HAS_WAITERS);
214 	}
215 }
216 
217 /*
218  * We can speed up the acquire/release, if there's no debugging state to be
219  * set up.
220  */
221 #ifndef CONFIG_DEBUG_RT_MUTEXES
222 static __always_inline bool rt_mutex_cmpxchg_acquire(struct rt_mutex_base *lock,
223 						     struct task_struct *old,
224 						     struct task_struct *new)
225 {
226 	return try_cmpxchg_acquire(&lock->owner, &old, new);
227 }
228 
229 static __always_inline bool rt_mutex_try_acquire(struct rt_mutex_base *lock)
230 {
231 	return rt_mutex_cmpxchg_acquire(lock, NULL, current);
232 }
233 
234 static __always_inline bool rt_mutex_cmpxchg_release(struct rt_mutex_base *lock,
235 						     struct task_struct *old,
236 						     struct task_struct *new)
237 {
238 	return try_cmpxchg_release(&lock->owner, &old, new);
239 }
240 
241 /*
242  * Callers must hold the ->wait_lock -- which is the whole purpose as we force
243  * all future threads that attempt to [Rmw] the lock to the slowpath. As such
244  * relaxed semantics suffice.
245  */
246 static __always_inline void mark_rt_mutex_waiters(struct rt_mutex_base *lock)
247 {
248 	unsigned long *p = (unsigned long *) &lock->owner;
249 	unsigned long owner, new;
250 
251 	owner = READ_ONCE(*p);
252 	do {
253 		new = owner | RT_MUTEX_HAS_WAITERS;
254 	} while (!try_cmpxchg_relaxed(p, &owner, new));
255 
256 	/*
257 	 * The cmpxchg loop above is relaxed to avoid back-to-back ACQUIRE
258 	 * operations in the event of contention. Ensure the successful
259 	 * cmpxchg is visible.
260 	 */
261 	smp_mb__after_atomic();
262 }
263 
264 /*
265  * Safe fastpath aware unlock:
266  * 1) Clear the waiters bit
267  * 2) Drop lock->wait_lock
268  * 3) Try to unlock the lock with cmpxchg
269  */
270 static __always_inline bool unlock_rt_mutex_safe(struct rt_mutex_base *lock,
271 						 unsigned long flags)
272 	__releases(lock->wait_lock)
273 {
274 	struct task_struct *owner = rt_mutex_owner(lock);
275 
276 	clear_rt_mutex_waiters(lock);
277 	raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
278 	/*
279 	 * If a new waiter comes in between the unlock and the cmpxchg
280 	 * we have two situations:
281 	 *
282 	 * unlock(wait_lock);
283 	 *					lock(wait_lock);
284 	 * cmpxchg(p, owner, 0) == owner
285 	 *					mark_rt_mutex_waiters(lock);
286 	 *					acquire(lock);
287 	 * or:
288 	 *
289 	 * unlock(wait_lock);
290 	 *					lock(wait_lock);
291 	 *					mark_rt_mutex_waiters(lock);
292 	 *
293 	 * cmpxchg(p, owner, 0) != owner
294 	 *					enqueue_waiter();
295 	 *					unlock(wait_lock);
296 	 * lock(wait_lock);
297 	 * wake waiter();
298 	 * unlock(wait_lock);
299 	 *					lock(wait_lock);
300 	 *					acquire(lock);
301 	 */
302 	return rt_mutex_cmpxchg_release(lock, owner, NULL);
303 }
304 
305 #else
306 static __always_inline bool rt_mutex_cmpxchg_acquire(struct rt_mutex_base *lock,
307 						     struct task_struct *old,
308 						     struct task_struct *new)
309 {
310 	return false;
311 
312 }
313 
314 static int __sched rt_mutex_slowtrylock(struct rt_mutex_base *lock);
315 
316 static __always_inline bool rt_mutex_try_acquire(struct rt_mutex_base *lock)
317 {
318 	/*
319 	 * With debug enabled rt_mutex_cmpxchg trylock() will always fail.
320 	 *
321 	 * Avoid unconditionally taking the slow path by using
322 	 * rt_mutex_slow_trylock() which is covered by the debug code and can
323 	 * acquire a non-contended rtmutex.
324 	 */
325 	return rt_mutex_slowtrylock(lock);
326 }
327 
328 static __always_inline bool rt_mutex_cmpxchg_release(struct rt_mutex_base *lock,
329 						     struct task_struct *old,
330 						     struct task_struct *new)
331 {
332 	return false;
333 }
334 
335 static __always_inline void mark_rt_mutex_waiters(struct rt_mutex_base *lock)
336 	__must_hold(&lock->wait_lock)
337 {
338 	lock->owner = (struct task_struct *)
339 			((unsigned long)lock->owner | RT_MUTEX_HAS_WAITERS);
340 }
341 
342 /*
343  * Simple slow path only version: lock->owner is protected by lock->wait_lock.
344  */
345 static __always_inline bool unlock_rt_mutex_safe(struct rt_mutex_base *lock,
346 						 unsigned long flags)
347 	__releases(lock->wait_lock)
348 {
349 	lock->owner = NULL;
350 	raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
351 	return true;
352 }
353 #endif
354 
355 static __always_inline int __waiter_prio(struct task_struct *task)
356 {
357 	int prio = task->prio;
358 
359 	if (!rt_or_dl_prio(prio))
360 		return DEFAULT_PRIO;
361 
362 	return prio;
363 }
364 
365 /*
366  * Update the waiter->tree copy of the sort keys.
367  */
368 static __always_inline void
369 waiter_update_prio(struct rt_mutex_waiter *waiter, struct task_struct *task)
370 {
371 	lockdep_assert_held(&waiter->lock->wait_lock);
372 	lockdep_assert(RB_EMPTY_NODE(&waiter->tree.entry));
373 
374 	waiter->tree.prio = __waiter_prio(task);
375 	waiter->tree.deadline = task->dl.deadline;
376 }
377 
378 /*
379  * Update the waiter->pi_tree copy of the sort keys (from the tree copy).
380  */
381 static __always_inline void
382 waiter_clone_prio(struct rt_mutex_waiter *waiter, struct task_struct *task)
383 {
384 	lockdep_assert_held(&waiter->lock->wait_lock);
385 	lockdep_assert_held(&task->pi_lock);
386 	lockdep_assert(RB_EMPTY_NODE(&waiter->pi_tree.entry));
387 
388 	waiter->pi_tree.prio = waiter->tree.prio;
389 	waiter->pi_tree.deadline = waiter->tree.deadline;
390 }
391 
392 /*
393  * Only use with rt_waiter_node_{less,equal}()
394  */
395 #define task_to_waiter_node(p)	\
396 	&(struct rt_waiter_node){ .prio = __waiter_prio(p), .deadline = (p)->dl.deadline }
397 #define task_to_waiter(p)	\
398 	&(struct rt_mutex_waiter){ .tree = *task_to_waiter_node(p) }
399 
400 static __always_inline int rt_waiter_node_less(struct rt_waiter_node *left,
401 					       struct rt_waiter_node *right)
402 {
403 	if (left->prio < right->prio)
404 		return 1;
405 
406 	/*
407 	 * If both waiters have dl_prio(), we check the deadlines of the
408 	 * associated tasks.
409 	 * If left waiter has a dl_prio(), and we didn't return 1 above,
410 	 * then right waiter has a dl_prio() too.
411 	 */
412 	if (dl_prio(left->prio))
413 		return dl_time_before(left->deadline, right->deadline);
414 
415 	return 0;
416 }
417 
418 static __always_inline int rt_waiter_node_equal(struct rt_waiter_node *left,
419 						 struct rt_waiter_node *right)
420 {
421 	if (left->prio != right->prio)
422 		return 0;
423 
424 	/*
425 	 * If both waiters have dl_prio(), we check the deadlines of the
426 	 * associated tasks.
427 	 * If left waiter has a dl_prio(), and we didn't return 0 above,
428 	 * then right waiter has a dl_prio() too.
429 	 */
430 	if (dl_prio(left->prio))
431 		return left->deadline == right->deadline;
432 
433 	return 1;
434 }
435 
436 static inline bool rt_mutex_steal(struct rt_mutex_waiter *waiter,
437 				  struct rt_mutex_waiter *top_waiter)
438 {
439 	if (rt_waiter_node_less(&waiter->tree, &top_waiter->tree))
440 		return true;
441 
442 #ifdef RT_MUTEX_BUILD_SPINLOCKS
443 	/*
444 	 * Note that RT tasks are excluded from same priority (lateral)
445 	 * steals to prevent the introduction of an unbounded latency.
446 	 */
447 	if (rt_or_dl_prio(waiter->tree.prio))
448 		return false;
449 
450 	return rt_waiter_node_equal(&waiter->tree, &top_waiter->tree);
451 #else
452 	return false;
453 #endif
454 }
455 
456 #define __node_2_waiter(node) \
457 	rb_entry((node), struct rt_mutex_waiter, tree.entry)
458 
459 static __always_inline bool __waiter_less(struct rb_node *a, const struct rb_node *b)
460 {
461 	struct rt_mutex_waiter *aw = __node_2_waiter(a);
462 	struct rt_mutex_waiter *bw = __node_2_waiter(b);
463 
464 	if (rt_waiter_node_less(&aw->tree, &bw->tree))
465 		return 1;
466 
467 	if (!build_ww_mutex())
468 		return 0;
469 
470 	if (rt_waiter_node_less(&bw->tree, &aw->tree))
471 		return 0;
472 
473 	/* NOTE: relies on waiter->ww_ctx being set before insertion */
474 	if (aw->ww_ctx) {
475 		if (!bw->ww_ctx)
476 			return 1;
477 
478 		return (signed long)(aw->ww_ctx->stamp -
479 				     bw->ww_ctx->stamp) < 0;
480 	}
481 
482 	return 0;
483 }
484 
485 static __always_inline void
486 rt_mutex_enqueue(struct rt_mutex_base *lock, struct rt_mutex_waiter *waiter)
487 	__must_hold(&lock->wait_lock)
488 {
489 	lockdep_assert_held(&lock->wait_lock);
490 
491 	rb_add_cached(&waiter->tree.entry, &lock->waiters, __waiter_less);
492 }
493 
494 static __always_inline void
495 rt_mutex_dequeue(struct rt_mutex_base *lock, struct rt_mutex_waiter *waiter)
496 	__must_hold(&lock->wait_lock)
497 {
498 	lockdep_assert_held(&lock->wait_lock);
499 
500 	if (RB_EMPTY_NODE(&waiter->tree.entry))
501 		return;
502 
503 	rb_erase_cached(&waiter->tree.entry, &lock->waiters);
504 	RB_CLEAR_NODE(&waiter->tree.entry);
505 }
506 
507 #define __node_2_rt_node(node) \
508 	rb_entry((node), struct rt_waiter_node, entry)
509 
510 static __always_inline bool __pi_waiter_less(struct rb_node *a, const struct rb_node *b)
511 {
512 	return rt_waiter_node_less(__node_2_rt_node(a), __node_2_rt_node(b));
513 }
514 
515 static __always_inline void
516 rt_mutex_enqueue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
517 {
518 	lockdep_assert_held(&task->pi_lock);
519 
520 	rb_add_cached(&waiter->pi_tree.entry, &task->pi_waiters, __pi_waiter_less);
521 }
522 
523 static __always_inline void
524 rt_mutex_dequeue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
525 {
526 	lockdep_assert_held(&task->pi_lock);
527 
528 	if (RB_EMPTY_NODE(&waiter->pi_tree.entry))
529 		return;
530 
531 	rb_erase_cached(&waiter->pi_tree.entry, &task->pi_waiters);
532 	RB_CLEAR_NODE(&waiter->pi_tree.entry);
533 }
534 
535 static __always_inline void rt_mutex_adjust_prio(struct rt_mutex_base *lock,
536 						 struct task_struct *p)
537 {
538 	struct task_struct *pi_task = NULL;
539 
540 	lockdep_assert_held(&lock->wait_lock);
541 	lockdep_assert(rt_mutex_owner(lock) == p);
542 	lockdep_assert_held(&p->pi_lock);
543 
544 	if (task_has_pi_waiters(p))
545 		pi_task = task_top_pi_waiter(p)->task;
546 
547 	rt_mutex_setprio(p, pi_task);
548 }
549 
550 /* RT mutex specific wake_q wrappers */
551 static __always_inline void rt_mutex_wake_q_add_task(struct rt_wake_q_head *wqh,
552 						     struct task_struct *task,
553 						     unsigned int wake_state)
554 {
555 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && wake_state == TASK_RTLOCK_WAIT) {
556 		if (IS_ENABLED(CONFIG_PROVE_LOCKING))
557 			WARN_ON_ONCE(wqh->rtlock_task);
558 		get_task_struct(task);
559 		wqh->rtlock_task = task;
560 	} else {
561 		wake_q_add(&wqh->head, task);
562 	}
563 }
564 
565 static __always_inline void rt_mutex_wake_q_add(struct rt_wake_q_head *wqh,
566 						struct rt_mutex_waiter *w)
567 {
568 	rt_mutex_wake_q_add_task(wqh, w->task, w->wake_state);
569 }
570 
571 static __always_inline void rt_mutex_wake_up_q(struct rt_wake_q_head *wqh)
572 {
573 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && wqh->rtlock_task) {
574 		wake_up_state(wqh->rtlock_task, TASK_RTLOCK_WAIT);
575 		put_task_struct(wqh->rtlock_task);
576 		wqh->rtlock_task = NULL;
577 	}
578 
579 	if (!wake_q_empty(&wqh->head))
580 		wake_up_q(&wqh->head);
581 
582 	/* Pairs with preempt_disable() in mark_wakeup_next_waiter() */
583 	preempt_enable();
584 }
585 
586 /*
587  * Deadlock detection is conditional:
588  *
589  * If CONFIG_DEBUG_RT_MUTEXES=n, deadlock detection is only conducted
590  * if the detect argument is == RT_MUTEX_FULL_CHAINWALK.
591  *
592  * If CONFIG_DEBUG_RT_MUTEXES=y, deadlock detection is always
593  * conducted independent of the detect argument.
594  *
595  * If the waiter argument is NULL this indicates the deboost path and
596  * deadlock detection is disabled independent of the detect argument
597  * and the config settings.
598  */
599 static __always_inline bool
600 rt_mutex_cond_detect_deadlock(struct rt_mutex_waiter *waiter,
601 			      enum rtmutex_chainwalk chwalk)
602 {
603 	if (IS_ENABLED(CONFIG_DEBUG_RT_MUTEXES))
604 		return waiter != NULL;
605 	return chwalk == RT_MUTEX_FULL_CHAINWALK;
606 }
607 
608 static __always_inline struct rt_mutex_base *task_blocked_on_lock(struct task_struct *p)
609 {
610 	return p->pi_blocked_on ? p->pi_blocked_on->lock : NULL;
611 }
612 
613 /*
614  * Adjust the priority chain. Also used for deadlock detection.
615  * Decreases task's usage by one - may thus free the task.
616  *
617  * @task:	the task owning the mutex (owner) for which a chain walk is
618  *		probably needed
619  * @chwalk:	do we have to carry out deadlock detection?
620  * @orig_lock:	the mutex (can be NULL if we are walking the chain to recheck
621  *		things for a task that has just got its priority adjusted, and
622  *		is waiting on a mutex)
623  * @next_lock:	the mutex on which the owner of @orig_lock was blocked before
624  *		we dropped its pi_lock. Is never dereferenced, only used for
625  *		comparison to detect lock chain changes.
626  * @orig_waiter: rt_mutex_waiter struct for the task that has just donated
627  *		its priority to the mutex owner (can be NULL in the case
628  *		depicted above or if the top waiter is gone away and we are
629  *		actually deboosting the owner)
630  * @top_task:	the current top waiter
631  *
632  * Returns 0 or -EDEADLK.
633  *
634  * Chain walk basics and protection scope
635  *
636  * [R] refcount on task
637  * [Pn] task->pi_lock held
638  * [L] rtmutex->wait_lock held
639  *
640  * Normal locking order:
641  *
642  *   rtmutex->wait_lock
643  *     task->pi_lock
644  *
645  * Step	Description				Protected by
646  *	function arguments:
647  *	@task					[R]
648  *	@orig_lock if != NULL			@top_task is blocked on it
649  *	@next_lock				Unprotected. Cannot be
650  *						dereferenced. Only used for
651  *						comparison.
652  *	@orig_waiter if != NULL			@top_task is blocked on it
653  *	@top_task				current, or in case of proxy
654  *						locking protected by calling
655  *						code
656  *	again:
657  *	  loop_sanity_check();
658  *	retry:
659  * [1]	  lock(task->pi_lock);			[R] acquire [P1]
660  * [2]	  waiter = task->pi_blocked_on;		[P1]
661  * [3]	  check_exit_conditions_1();		[P1]
662  * [4]	  lock = waiter->lock;			[P1]
663  * [5]	  if (!try_lock(lock->wait_lock)) {	[P1] try to acquire [L]
664  *	    unlock(task->pi_lock);		release [P1]
665  *	    goto retry;
666  *	  }
667  * [6]	  check_exit_conditions_2();		[P1] + [L]
668  * [7]	  requeue_lock_waiter(lock, waiter);	[P1] + [L]
669  * [8]	  unlock(task->pi_lock);		release [P1]
670  *	  put_task_struct(task);		release [R]
671  * [9]	  check_exit_conditions_3();		[L]
672  * [10]	  task = owner(lock);			[L]
673  *	  get_task_struct(task);		[L] acquire [R]
674  *	  lock(task->pi_lock);			[L] acquire [P2]
675  * [11]	  requeue_pi_waiter(tsk, waiters(lock));[P2] + [L]
676  * [12]	  check_exit_conditions_4();		[P2] + [L]
677  * [13]	  unlock(task->pi_lock);		release [P2]
678  *	  unlock(lock->wait_lock);		release [L]
679  *	  goto again;
680  *
681  * Where P1 is the blocking task and P2 is the lock owner; going up one step
682  * the owner becomes the next blocked task etc..
683  *
684 *
685  */
686 static int __sched rt_mutex_adjust_prio_chain(struct task_struct *task,
687 					      enum rtmutex_chainwalk chwalk,
688 					      struct rt_mutex_base *orig_lock,
689 					      struct rt_mutex_base *next_lock,
690 					      struct rt_mutex_waiter *orig_waiter,
691 					      struct task_struct *top_task)
692 {
693 	struct rt_mutex_waiter *waiter, *top_waiter = orig_waiter;
694 	struct rt_mutex_waiter *prerequeue_top_waiter;
695 	int ret = 0, depth = 0;
696 	struct rt_mutex_base *lock;
697 	bool detect_deadlock;
698 	bool requeue = true;
699 
700 	detect_deadlock = rt_mutex_cond_detect_deadlock(orig_waiter, chwalk);
701 
702 	/*
703 	 * The (de)boosting is a step by step approach with a lot of
704 	 * pitfalls. We want this to be preemptible and we want hold a
705 	 * maximum of two locks per step. So we have to check
706 	 * carefully whether things change under us.
707 	 */
708  again:
709 	/*
710 	 * We limit the lock chain length for each invocation.
711 	 */
712 	if (++depth > max_lock_depth) {
713 		static int prev_max;
714 
715 		/*
716 		 * Print this only once. If the admin changes the limit,
717 		 * print a new message when reaching the limit again.
718 		 */
719 		if (prev_max != max_lock_depth) {
720 			prev_max = max_lock_depth;
721 			printk(KERN_WARNING "Maximum lock depth %d reached "
722 			       "task: %s (%d)\n", max_lock_depth,
723 			       top_task->comm, task_pid_nr(top_task));
724 		}
725 		put_task_struct(task);
726 
727 		return -EDEADLK;
728 	}
729 
730 	/*
731 	 * We are fully preemptible here and only hold the refcount on
732 	 * @task. So everything can have changed under us since the
733 	 * caller or our own code below (goto retry/again) dropped all
734 	 * locks.
735 	 */
736  retry:
737 	/*
738 	 * [1] Task cannot go away as we did a get_task() before !
739 	 */
740 	raw_spin_lock_irq(&task->pi_lock);
741 
742 	/*
743 	 * [2] Get the waiter on which @task is blocked on.
744 	 */
745 	waiter = task->pi_blocked_on;
746 
747 	/*
748 	 * [3] check_exit_conditions_1() protected by task->pi_lock.
749 	 */
750 
751 	/*
752 	 * Check whether the end of the boosting chain has been
753 	 * reached or the state of the chain has changed while we
754 	 * dropped the locks.
755 	 */
756 	if (!waiter)
757 		goto out_unlock_pi;
758 
759 	/*
760 	 * Check the orig_waiter state. After we dropped the locks,
761 	 * the previous owner of the lock might have released the lock.
762 	 */
763 	if (orig_waiter && !rt_mutex_owner(orig_lock))
764 		goto out_unlock_pi;
765 
766 	/*
767 	 * We dropped all locks after taking a refcount on @task, so
768 	 * the task might have moved on in the lock chain or even left
769 	 * the chain completely and blocks now on an unrelated lock or
770 	 * on @orig_lock.
771 	 *
772 	 * We stored the lock on which @task was blocked in @next_lock,
773 	 * so we can detect the chain change.
774 	 */
775 	if (next_lock != waiter->lock)
776 		goto out_unlock_pi;
777 
778 	/*
779 	 * There could be 'spurious' loops in the lock graph due to ww_mutex,
780 	 * consider:
781 	 *
782 	 *   P1: A, ww_A, ww_B
783 	 *   P2: ww_B, ww_A
784 	 *   P3: A
785 	 *
786 	 * P3 should not return -EDEADLK because it gets trapped in the cycle
787 	 * created by P1 and P2 (which will resolve -- and runs into
788 	 * max_lock_depth above). Therefore disable detect_deadlock such that
789 	 * the below termination condition can trigger once all relevant tasks
790 	 * are boosted.
791 	 *
792 	 * Even when we start with ww_mutex we can disable deadlock detection,
793 	 * since we would supress a ww_mutex induced deadlock at [6] anyway.
794 	 * Supressing it here however is not sufficient since we might still
795 	 * hit [6] due to adjustment driven iteration.
796 	 *
797 	 * NOTE: if someone were to create a deadlock between 2 ww_classes we'd
798 	 * utterly fail to report it; lockdep should.
799 	 */
800 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && waiter->ww_ctx && detect_deadlock)
801 		detect_deadlock = false;
802 
803 	/*
804 	 * Drop out, when the task has no waiters. Note,
805 	 * top_waiter can be NULL, when we are in the deboosting
806 	 * mode!
807 	 */
808 	if (top_waiter) {
809 		if (!task_has_pi_waiters(task))
810 			goto out_unlock_pi;
811 		/*
812 		 * If deadlock detection is off, we stop here if we
813 		 * are not the top pi waiter of the task. If deadlock
814 		 * detection is enabled we continue, but stop the
815 		 * requeueing in the chain walk.
816 		 */
817 		if (top_waiter != task_top_pi_waiter(task)) {
818 			if (!detect_deadlock)
819 				goto out_unlock_pi;
820 			else
821 				requeue = false;
822 		}
823 	}
824 
825 	/*
826 	 * If the waiter priority is the same as the task priority
827 	 * then there is no further priority adjustment necessary.  If
828 	 * deadlock detection is off, we stop the chain walk. If its
829 	 * enabled we continue, but stop the requeueing in the chain
830 	 * walk.
831 	 */
832 	if (rt_waiter_node_equal(&waiter->tree, task_to_waiter_node(task))) {
833 		if (!detect_deadlock)
834 			goto out_unlock_pi;
835 		else
836 			requeue = false;
837 	}
838 
839 	/*
840 	 * [4] Get the next lock; per holding task->pi_lock we can't unblock
841 	 * and guarantee @lock's existence.
842 	 */
843 	lock = waiter->lock;
844 	/*
845 	 * [5] We need to trylock here as we are holding task->pi_lock,
846 	 * which is the reverse lock order versus the other rtmutex
847 	 * operations.
848 	 *
849 	 * Per the above, holding task->pi_lock guarantees lock exists, so
850 	 * inverting this lock order is infeasible from a life-time
851 	 * perspective.
852 	 */
853 	if (!raw_spin_trylock(&lock->wait_lock)) {
854 		raw_spin_unlock_irq(&task->pi_lock);
855 		cpu_relax();
856 		goto retry;
857 	}
858 
859 	/*
860 	 * [6] check_exit_conditions_2() protected by task->pi_lock and
861 	 * lock->wait_lock.
862 	 *
863 	 * Deadlock detection. If the lock is the same as the original
864 	 * lock which caused us to walk the lock chain or if the
865 	 * current lock is owned by the task which initiated the chain
866 	 * walk, we detected a deadlock.
867 	 */
868 	if (lock == orig_lock || rt_mutex_owner(lock) == top_task) {
869 		ret = -EDEADLK;
870 
871 		/*
872 		 * When the deadlock is due to ww_mutex; also see above. Don't
873 		 * report the deadlock and instead let the ww_mutex wound/die
874 		 * logic pick which of the contending threads gets -EDEADLK.
875 		 *
876 		 * NOTE: assumes the cycle only contains a single ww_class; any
877 		 * other configuration and we fail to report; also, see
878 		 * lockdep.
879 		 */
880 		if (IS_ENABLED(CONFIG_PREEMPT_RT) && orig_waiter && orig_waiter->ww_ctx)
881 			ret = 0;
882 
883 		raw_spin_unlock(&lock->wait_lock);
884 		goto out_unlock_pi;
885 	}
886 
887 	/*
888 	 * If we just follow the lock chain for deadlock detection, no
889 	 * need to do all the requeue operations. To avoid a truckload
890 	 * of conditionals around the various places below, just do the
891 	 * minimum chain walk checks.
892 	 */
893 	if (!requeue) {
894 		/*
895 		 * No requeue[7] here. Just release @task [8]
896 		 */
897 		raw_spin_unlock(&task->pi_lock);
898 		put_task_struct(task);
899 
900 		/*
901 		 * [9] check_exit_conditions_3 protected by lock->wait_lock.
902 		 * If there is no owner of the lock, end of chain.
903 		 */
904 		if (!rt_mutex_owner(lock)) {
905 			raw_spin_unlock_irq(&lock->wait_lock);
906 			return 0;
907 		}
908 
909 		/* [10] Grab the next task, i.e. owner of @lock */
910 		task = get_task_struct(rt_mutex_owner(lock));
911 		raw_spin_lock(&task->pi_lock);
912 
913 		/*
914 		 * No requeue [11] here. We just do deadlock detection.
915 		 *
916 		 * [12] Store whether owner is blocked
917 		 * itself. Decision is made after dropping the locks
918 		 */
919 		next_lock = task_blocked_on_lock(task);
920 		/*
921 		 * Get the top waiter for the next iteration
922 		 */
923 		top_waiter = rt_mutex_top_waiter(lock);
924 
925 		/* [13] Drop locks */
926 		raw_spin_unlock(&task->pi_lock);
927 		raw_spin_unlock_irq(&lock->wait_lock);
928 
929 		/* If owner is not blocked, end of chain. */
930 		if (!next_lock)
931 			goto out_put_task;
932 		goto again;
933 	}
934 
935 	/*
936 	 * Store the current top waiter before doing the requeue
937 	 * operation on @lock. We need it for the boost/deboost
938 	 * decision below.
939 	 */
940 	prerequeue_top_waiter = rt_mutex_top_waiter(lock);
941 
942 	/* [7] Requeue the waiter in the lock waiter tree. */
943 	rt_mutex_dequeue(lock, waiter);
944 
945 	/*
946 	 * Update the waiter prio fields now that we're dequeued.
947 	 *
948 	 * These values can have changed through either:
949 	 *
950 	 *   sys_sched_set_scheduler() / sys_sched_setattr()
951 	 *
952 	 * or
953 	 *
954 	 *   DL CBS enforcement advancing the effective deadline.
955 	 */
956 	waiter_update_prio(waiter, task);
957 
958 	rt_mutex_enqueue(lock, waiter);
959 
960 	/*
961 	 * [8] Release the (blocking) task in preparation for
962 	 * taking the owner task in [10].
963 	 *
964 	 * Since we hold lock->waiter_lock, task cannot unblock, even if we
965 	 * release task->pi_lock.
966 	 */
967 	raw_spin_unlock(&task->pi_lock);
968 	put_task_struct(task);
969 
970 	/*
971 	 * [9] check_exit_conditions_3 protected by lock->wait_lock.
972 	 *
973 	 * We must abort the chain walk if there is no lock owner even
974 	 * in the dead lock detection case, as we have nothing to
975 	 * follow here. This is the end of the chain we are walking.
976 	 */
977 	if (!rt_mutex_owner(lock)) {
978 		/*
979 		 * If the requeue [7] above changed the top waiter,
980 		 * then we need to wake the new top waiter up to try
981 		 * to get the lock.
982 		 */
983 		top_waiter = rt_mutex_top_waiter(lock);
984 		if (prerequeue_top_waiter != top_waiter)
985 			wake_up_state(top_waiter->task, top_waiter->wake_state);
986 		raw_spin_unlock_irq(&lock->wait_lock);
987 		return 0;
988 	}
989 
990 	/*
991 	 * [10] Grab the next task, i.e. the owner of @lock
992 	 *
993 	 * Per holding lock->wait_lock and checking for !owner above, there
994 	 * must be an owner and it cannot go away.
995 	 */
996 	task = get_task_struct(rt_mutex_owner(lock));
997 	raw_spin_lock(&task->pi_lock);
998 
999 	/* [11] requeue the pi waiters if necessary */
1000 	if (waiter == rt_mutex_top_waiter(lock)) {
1001 		/*
1002 		 * The waiter became the new top (highest priority)
1003 		 * waiter on the lock. Replace the previous top waiter
1004 		 * in the owner tasks pi waiters tree with this waiter
1005 		 * and adjust the priority of the owner.
1006 		 */
1007 		rt_mutex_dequeue_pi(task, prerequeue_top_waiter);
1008 		waiter_clone_prio(waiter, task);
1009 		rt_mutex_enqueue_pi(task, waiter);
1010 		rt_mutex_adjust_prio(lock, task);
1011 
1012 	} else if (prerequeue_top_waiter == waiter) {
1013 		/*
1014 		 * The waiter was the top waiter on the lock, but is
1015 		 * no longer the top priority waiter. Replace waiter in
1016 		 * the owner tasks pi waiters tree with the new top
1017 		 * (highest priority) waiter and adjust the priority
1018 		 * of the owner.
1019 		 * The new top waiter is stored in @waiter so that
1020 		 * @waiter == @top_waiter evaluates to true below and
1021 		 * we continue to deboost the rest of the chain.
1022 		 */
1023 		rt_mutex_dequeue_pi(task, waiter);
1024 		waiter = rt_mutex_top_waiter(lock);
1025 		waiter_clone_prio(waiter, task);
1026 		rt_mutex_enqueue_pi(task, waiter);
1027 		rt_mutex_adjust_prio(lock, task);
1028 	} else {
1029 		/*
1030 		 * Nothing changed. No need to do any priority
1031 		 * adjustment.
1032 		 */
1033 	}
1034 
1035 	/*
1036 	 * [12] check_exit_conditions_4() protected by task->pi_lock
1037 	 * and lock->wait_lock. The actual decisions are made after we
1038 	 * dropped the locks.
1039 	 *
1040 	 * Check whether the task which owns the current lock is pi
1041 	 * blocked itself. If yes we store a pointer to the lock for
1042 	 * the lock chain change detection above. After we dropped
1043 	 * task->pi_lock next_lock cannot be dereferenced anymore.
1044 	 */
1045 	next_lock = task_blocked_on_lock(task);
1046 	/*
1047 	 * Store the top waiter of @lock for the end of chain walk
1048 	 * decision below.
1049 	 */
1050 	top_waiter = rt_mutex_top_waiter(lock);
1051 
1052 	/* [13] Drop the locks */
1053 	raw_spin_unlock(&task->pi_lock);
1054 	raw_spin_unlock_irq(&lock->wait_lock);
1055 
1056 	/*
1057 	 * Make the actual exit decisions [12], based on the stored
1058 	 * values.
1059 	 *
1060 	 * We reached the end of the lock chain. Stop right here. No
1061 	 * point to go back just to figure that out.
1062 	 */
1063 	if (!next_lock)
1064 		goto out_put_task;
1065 
1066 	/*
1067 	 * If the current waiter is not the top waiter on the lock,
1068 	 * then we can stop the chain walk here if we are not in full
1069 	 * deadlock detection mode.
1070 	 */
1071 	if (!detect_deadlock && waiter != top_waiter)
1072 		goto out_put_task;
1073 
1074 	goto again;
1075 
1076  out_unlock_pi:
1077 	raw_spin_unlock_irq(&task->pi_lock);
1078  out_put_task:
1079 	put_task_struct(task);
1080 
1081 	return ret;
1082 }
1083 
1084 /*
1085  * Try to take an rt-mutex
1086  *
1087  * Must be called with lock->wait_lock held and interrupts disabled
1088  *
1089  * @lock:   The lock to be acquired.
1090  * @task:   The task which wants to acquire the lock
1091  * @waiter: The waiter that is queued to the lock's wait tree if the
1092  *	    callsite called task_blocked_on_lock(), otherwise NULL
1093  */
1094 static int __sched
1095 try_to_take_rt_mutex(struct rt_mutex_base *lock, struct task_struct *task,
1096 		     struct rt_mutex_waiter *waiter)
1097 	__must_hold(&lock->wait_lock)
1098 {
1099 	lockdep_assert_held(&lock->wait_lock);
1100 
1101 	/*
1102 	 * Before testing whether we can acquire @lock, we set the
1103 	 * RT_MUTEX_HAS_WAITERS bit in @lock->owner. This forces all
1104 	 * other tasks which try to modify @lock into the slow path
1105 	 * and they serialize on @lock->wait_lock.
1106 	 *
1107 	 * The RT_MUTEX_HAS_WAITERS bit can have a transitional state
1108 	 * as explained at the top of this file if and only if:
1109 	 *
1110 	 * - There is a lock owner. The caller must fixup the
1111 	 *   transient state if it does a trylock or leaves the lock
1112 	 *   function due to a signal or timeout.
1113 	 *
1114 	 * - @task acquires the lock and there are no other
1115 	 *   waiters. This is undone in rt_mutex_set_owner(@task) at
1116 	 *   the end of this function.
1117 	 */
1118 	mark_rt_mutex_waiters(lock);
1119 
1120 	/*
1121 	 * If @lock has an owner, give up.
1122 	 */
1123 	if (rt_mutex_owner(lock))
1124 		return 0;
1125 
1126 	/*
1127 	 * If @waiter != NULL, @task has already enqueued the waiter
1128 	 * into @lock waiter tree. If @waiter == NULL then this is a
1129 	 * trylock attempt.
1130 	 */
1131 	if (waiter) {
1132 		struct rt_mutex_waiter *top_waiter = rt_mutex_top_waiter(lock);
1133 
1134 		/*
1135 		 * If waiter is the highest priority waiter of @lock,
1136 		 * or allowed to steal it, take it over.
1137 		 */
1138 		if (waiter == top_waiter || rt_mutex_steal(waiter, top_waiter)) {
1139 			/*
1140 			 * We can acquire the lock. Remove the waiter from the
1141 			 * lock waiters tree.
1142 			 */
1143 			rt_mutex_dequeue(lock, waiter);
1144 		} else {
1145 			return 0;
1146 		}
1147 	} else {
1148 		/*
1149 		 * If the lock has waiters already we check whether @task is
1150 		 * eligible to take over the lock.
1151 		 *
1152 		 * If there are no other waiters, @task can acquire
1153 		 * the lock.  @task->pi_blocked_on is NULL, so it does
1154 		 * not need to be dequeued.
1155 		 */
1156 		if (rt_mutex_has_waiters(lock)) {
1157 			/* Check whether the trylock can steal it. */
1158 			if (!rt_mutex_steal(task_to_waiter(task),
1159 					    rt_mutex_top_waiter(lock)))
1160 				return 0;
1161 
1162 			/*
1163 			 * The current top waiter stays enqueued. We
1164 			 * don't have to change anything in the lock
1165 			 * waiters order.
1166 			 */
1167 		} else {
1168 			/*
1169 			 * No waiters. Take the lock without the
1170 			 * pi_lock dance.@task->pi_blocked_on is NULL
1171 			 * and we have no waiters to enqueue in @task
1172 			 * pi waiters tree.
1173 			 */
1174 			goto takeit;
1175 		}
1176 	}
1177 
1178 	/*
1179 	 * Clear @task->pi_blocked_on. Requires protection by
1180 	 * @task->pi_lock. Redundant operation for the @waiter == NULL
1181 	 * case, but conditionals are more expensive than a redundant
1182 	 * store.
1183 	 */
1184 	raw_spin_lock(&task->pi_lock);
1185 	task->pi_blocked_on = NULL;
1186 	/*
1187 	 * Finish the lock acquisition. @task is the new owner. If
1188 	 * other waiters exist we have to insert the highest priority
1189 	 * waiter into @task->pi_waiters tree.
1190 	 */
1191 	if (rt_mutex_has_waiters(lock))
1192 		rt_mutex_enqueue_pi(task, rt_mutex_top_waiter(lock));
1193 	raw_spin_unlock(&task->pi_lock);
1194 
1195 takeit:
1196 	/*
1197 	 * This either preserves the RT_MUTEX_HAS_WAITERS bit if there
1198 	 * are still waiters or clears it.
1199 	 */
1200 	rt_mutex_set_owner(lock, task);
1201 
1202 	return 1;
1203 }
1204 
1205 /*
1206  * Task blocks on lock.
1207  *
1208  * Prepare waiter and propagate pi chain
1209  *
1210  * This must be called with lock->wait_lock held and interrupts disabled
1211  */
1212 static int __sched task_blocks_on_rt_mutex(struct rt_mutex_base *lock,
1213 					   struct rt_mutex_waiter *waiter,
1214 					   struct task_struct *task,
1215 					   struct ww_acquire_ctx *ww_ctx,
1216 					   enum rtmutex_chainwalk chwalk,
1217 					   struct wake_q_head *wake_q)
1218 	__must_hold(&lock->wait_lock)
1219 {
1220 	struct task_struct *owner = rt_mutex_owner(lock);
1221 	struct rt_mutex_waiter *top_waiter = waiter;
1222 	struct rt_mutex_base *next_lock;
1223 	int chain_walk = 0, res;
1224 
1225 	lockdep_assert_held(&lock->wait_lock);
1226 
1227 	/*
1228 	 * Early deadlock detection. We really don't want the task to
1229 	 * enqueue on itself just to untangle the mess later. It's not
1230 	 * only an optimization. We drop the locks, so another waiter
1231 	 * can come in before the chain walk detects the deadlock. So
1232 	 * the other will detect the deadlock and return -EDEADLOCK,
1233 	 * which is wrong, as the other waiter is not in a deadlock
1234 	 * situation.
1235 	 *
1236 	 * Except for ww_mutex, in that case the chain walk must already deal
1237 	 * with spurious cycles, see the comments at [3] and [6].
1238 	 */
1239 	if (owner == task && !(build_ww_mutex() && ww_ctx))
1240 		return -EDEADLK;
1241 
1242 	raw_spin_lock(&task->pi_lock);
1243 	waiter->task = task;
1244 	waiter->lock = lock;
1245 	waiter_update_prio(waiter, task);
1246 	waiter_clone_prio(waiter, task);
1247 
1248 	/* Get the top priority waiter on the lock */
1249 	if (rt_mutex_has_waiters(lock))
1250 		top_waiter = rt_mutex_top_waiter(lock);
1251 	rt_mutex_enqueue(lock, waiter);
1252 
1253 	task->pi_blocked_on = waiter;
1254 
1255 	raw_spin_unlock(&task->pi_lock);
1256 
1257 	if (build_ww_mutex() && ww_ctx) {
1258 		struct rt_mutex *rtm;
1259 
1260 		/* Check whether the waiter should back out immediately */
1261 		rtm = container_of(lock, struct rt_mutex, rtmutex);
1262 		__assume_ctx_lock(&rtm->rtmutex.wait_lock);
1263 		res = __ww_mutex_add_waiter(waiter, rtm, ww_ctx, wake_q);
1264 		if (res) {
1265 			raw_spin_lock(&task->pi_lock);
1266 			rt_mutex_dequeue(lock, waiter);
1267 			task->pi_blocked_on = NULL;
1268 			raw_spin_unlock(&task->pi_lock);
1269 			return res;
1270 		}
1271 	}
1272 
1273 	if (!owner)
1274 		return 0;
1275 
1276 	raw_spin_lock(&owner->pi_lock);
1277 	if (waiter == rt_mutex_top_waiter(lock)) {
1278 		rt_mutex_dequeue_pi(owner, top_waiter);
1279 		rt_mutex_enqueue_pi(owner, waiter);
1280 
1281 		rt_mutex_adjust_prio(lock, owner);
1282 		if (owner->pi_blocked_on)
1283 			chain_walk = 1;
1284 	} else if (rt_mutex_cond_detect_deadlock(waiter, chwalk)) {
1285 		chain_walk = 1;
1286 	}
1287 
1288 	/* Store the lock on which owner is blocked or NULL */
1289 	next_lock = task_blocked_on_lock(owner);
1290 
1291 	raw_spin_unlock(&owner->pi_lock);
1292 	/*
1293 	 * Even if full deadlock detection is on, if the owner is not
1294 	 * blocked itself, we can avoid finding this out in the chain
1295 	 * walk.
1296 	 */
1297 	if (!chain_walk || !next_lock)
1298 		return 0;
1299 
1300 	/*
1301 	 * The owner can't disappear while holding a lock,
1302 	 * so the owner struct is protected by wait_lock.
1303 	 * Gets dropped in rt_mutex_adjust_prio_chain()!
1304 	 */
1305 	get_task_struct(owner);
1306 
1307 	raw_spin_unlock_irq_wake(&lock->wait_lock, wake_q);
1308 
1309 	res = rt_mutex_adjust_prio_chain(owner, chwalk, lock,
1310 					 next_lock, waiter, task);
1311 
1312 	raw_spin_lock_irq(&lock->wait_lock);
1313 
1314 	return res;
1315 }
1316 
1317 /*
1318  * Remove the top waiter from the current tasks pi waiter tree and
1319  * queue it up.
1320  *
1321  * Called with lock->wait_lock held and interrupts disabled.
1322  */
1323 static void __sched mark_wakeup_next_waiter(struct rt_wake_q_head *wqh,
1324 					    struct rt_mutex_base *lock)
1325 	__must_hold(&lock->wait_lock)
1326 {
1327 	struct rt_mutex_waiter *waiter;
1328 
1329 	lockdep_assert_held(&lock->wait_lock);
1330 
1331 	raw_spin_lock(&current->pi_lock);
1332 
1333 	waiter = rt_mutex_top_waiter(lock);
1334 
1335 	/*
1336 	 * Remove it from current->pi_waiters and deboost.
1337 	 *
1338 	 * We must in fact deboost here in order to ensure we call
1339 	 * rt_mutex_setprio() to update p->pi_top_task before the
1340 	 * task unblocks.
1341 	 */
1342 	rt_mutex_dequeue_pi(current, waiter);
1343 	rt_mutex_adjust_prio(lock, current);
1344 
1345 	/*
1346 	 * As we are waking up the top waiter, and the waiter stays
1347 	 * queued on the lock until it gets the lock, this lock
1348 	 * obviously has waiters. Just set the bit here and this has
1349 	 * the added benefit of forcing all new tasks into the
1350 	 * slow path making sure no task of lower priority than
1351 	 * the top waiter can steal this lock.
1352 	 */
1353 	lock->owner = (void *) RT_MUTEX_HAS_WAITERS;
1354 
1355 	/*
1356 	 * We deboosted before waking the top waiter task such that we don't
1357 	 * run two tasks with the 'same' priority (and ensure the
1358 	 * p->pi_top_task pointer points to a blocked task). This however can
1359 	 * lead to priority inversion if we would get preempted after the
1360 	 * deboost but before waking our donor task, hence the preempt_disable()
1361 	 * before unlock.
1362 	 *
1363 	 * Pairs with preempt_enable() in rt_mutex_wake_up_q();
1364 	 */
1365 	preempt_disable();
1366 	rt_mutex_wake_q_add(wqh, waiter);
1367 	raw_spin_unlock(&current->pi_lock);
1368 }
1369 
1370 static int __sched __rt_mutex_slowtrylock(struct rt_mutex_base *lock)
1371 	__must_hold(&lock->wait_lock)
1372 {
1373 	int ret = try_to_take_rt_mutex(lock, current, NULL);
1374 
1375 	/*
1376 	 * try_to_take_rt_mutex() sets the lock waiters bit
1377 	 * unconditionally. Clean this up.
1378 	 */
1379 	fixup_rt_mutex_waiters(lock, true);
1380 
1381 	return ret;
1382 }
1383 
1384 /*
1385  * Slow path try-lock function:
1386  */
1387 static int __sched rt_mutex_slowtrylock(struct rt_mutex_base *lock)
1388 {
1389 	unsigned long flags;
1390 	int ret;
1391 
1392 	/*
1393 	 * If the lock already has an owner we fail to get the lock.
1394 	 * This can be done without taking the @lock->wait_lock as
1395 	 * it is only being read, and this is a trylock anyway.
1396 	 */
1397 	if (rt_mutex_owner(lock))
1398 		return 0;
1399 
1400 	/*
1401 	 * The mutex has currently no owner. Lock the wait lock and try to
1402 	 * acquire the lock. We use irqsave here to support early boot calls.
1403 	 */
1404 	raw_spin_lock_irqsave(&lock->wait_lock, flags);
1405 
1406 	ret = __rt_mutex_slowtrylock(lock);
1407 
1408 	raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1409 
1410 	return ret;
1411 }
1412 
1413 static __always_inline int __rt_mutex_trylock(struct rt_mutex_base *lock)
1414 {
1415 	if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
1416 		return 1;
1417 
1418 	return rt_mutex_slowtrylock(lock);
1419 }
1420 
1421 /*
1422  * Slow path to release a rt-mutex.
1423  */
1424 static void __sched rt_mutex_slowunlock(struct rt_mutex_base *lock)
1425 {
1426 	DEFINE_RT_WAKE_Q(wqh);
1427 	unsigned long flags;
1428 
1429 	/* irqsave required to support early boot calls */
1430 	raw_spin_lock_irqsave(&lock->wait_lock, flags);
1431 
1432 	debug_rt_mutex_unlock(lock);
1433 
1434 	/*
1435 	 * We must be careful here if the fast path is enabled. If we
1436 	 * have no waiters queued we cannot set owner to NULL here
1437 	 * because of:
1438 	 *
1439 	 * foo->lock->owner = NULL;
1440 	 *			rtmutex_lock(foo->lock);   <- fast path
1441 	 *			free = atomic_dec_and_test(foo->refcnt);
1442 	 *			rtmutex_unlock(foo->lock); <- fast path
1443 	 *			if (free)
1444 	 *				kfree(foo);
1445 	 * raw_spin_unlock(foo->lock->wait_lock);
1446 	 *
1447 	 * So for the fastpath enabled kernel:
1448 	 *
1449 	 * Nothing can set the waiters bit as long as we hold
1450 	 * lock->wait_lock. So we do the following sequence:
1451 	 *
1452 	 *	owner = rt_mutex_owner(lock);
1453 	 *	clear_rt_mutex_waiters(lock);
1454 	 *	raw_spin_unlock(&lock->wait_lock);
1455 	 *	if (cmpxchg(&lock->owner, owner, 0) == owner)
1456 	 *		return;
1457 	 *	goto retry;
1458 	 *
1459 	 * The fastpath disabled variant is simple as all access to
1460 	 * lock->owner is serialized by lock->wait_lock:
1461 	 *
1462 	 *	lock->owner = NULL;
1463 	 *	raw_spin_unlock(&lock->wait_lock);
1464 	 */
1465 	while (!rt_mutex_has_waiters(lock)) {
1466 		/* Drops lock->wait_lock ! */
1467 		if (unlock_rt_mutex_safe(lock, flags) == true)
1468 			return;
1469 		/* Relock the rtmutex and try again */
1470 		raw_spin_lock_irqsave(&lock->wait_lock, flags);
1471 	}
1472 
1473 	trace_contended_release(lock);
1474 	/*
1475 	 * The wakeup next waiter path does not suffer from the above
1476 	 * race. See the comments there.
1477 	 *
1478 	 * Queue the next waiter for wakeup once we release the wait_lock.
1479 	 */
1480 	mark_wakeup_next_waiter(&wqh, lock);
1481 	raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1482 
1483 	rt_mutex_wake_up_q(&wqh);
1484 }
1485 
1486 static __always_inline void __rt_mutex_unlock(struct rt_mutex_base *lock)
1487 {
1488 	if (likely(rt_mutex_cmpxchg_release(lock, current, NULL)))
1489 		return;
1490 
1491 	rt_mutex_slowunlock(lock);
1492 }
1493 
1494 #ifdef CONFIG_SMP
1495 static bool rtmutex_spin_on_owner(struct rt_mutex_base *lock,
1496 				  struct rt_mutex_waiter *waiter,
1497 				  struct task_struct *owner)
1498 {
1499 	bool res = true;
1500 
1501 	rcu_read_lock();
1502 	for (;;) {
1503 		/* If owner changed, trylock again. */
1504 		if (owner != rt_mutex_owner(lock))
1505 			break;
1506 		/*
1507 		 * Ensure that @owner is dereferenced after checking that
1508 		 * the lock owner still matches @owner. If that fails,
1509 		 * @owner might point to freed memory. If it still matches,
1510 		 * the rcu_read_lock() ensures the memory stays valid.
1511 		 */
1512 		barrier();
1513 		/*
1514 		 * Stop spinning when:
1515 		 *  - the lock owner has been scheduled out
1516 		 *  - current is not longer the top waiter
1517 		 *  - current is requested to reschedule (redundant
1518 		 *    for CONFIG_PREEMPT_RCU=y)
1519 		 *  - the VCPU on which owner runs is preempted
1520 		 */
1521 		if (!owner_on_cpu(owner) || need_resched() ||
1522 		    !data_race(rt_mutex_waiter_is_top_waiter(lock, waiter))) {
1523 			res = false;
1524 			break;
1525 		}
1526 		cpu_relax();
1527 	}
1528 	rcu_read_unlock();
1529 	return res;
1530 }
1531 #else
1532 static bool rtmutex_spin_on_owner(struct rt_mutex_base *lock,
1533 				  struct rt_mutex_waiter *waiter,
1534 				  struct task_struct *owner)
1535 {
1536 	return false;
1537 }
1538 #endif
1539 
1540 #ifdef RT_MUTEX_BUILD_MUTEX
1541 /*
1542  * Functions required for:
1543  *	- rtmutex, futex on all kernels
1544  *	- mutex and rwsem substitutions on RT kernels
1545  */
1546 
1547 /*
1548  * Remove a waiter from a lock and give up
1549  *
1550  * Must be called with lock->wait_lock held and interrupts disabled. It must
1551  * have just failed to try_to_take_rt_mutex().
1552  *
1553  * When invoked from rt_mutex_start_proxy_lock() waiter::task != current !
1554  */
1555 static void __sched remove_waiter(struct rt_mutex_base *lock,
1556 				  struct rt_mutex_waiter *waiter)
1557 	__must_hold(&lock->wait_lock)
1558 {
1559 	bool is_top_waiter = (waiter == rt_mutex_top_waiter(lock));
1560 	struct task_struct *owner = rt_mutex_owner(lock);
1561 	struct task_struct *waiter_task = waiter->task;
1562 	struct rt_mutex_base *next_lock;
1563 
1564 	lockdep_assert_held(&lock->wait_lock);
1565 
1566 	if (!waiter_task) /* never enqueued */
1567 		return;
1568 
1569 	scoped_guard(raw_spinlock, &waiter_task->pi_lock) {
1570 		rt_mutex_dequeue(lock, waiter);
1571 		waiter_task->pi_blocked_on = NULL;
1572 	}
1573 
1574 	/*
1575 	 * Only update priority if the waiter was the highest priority
1576 	 * waiter of the lock and there is an owner to update.
1577 	 */
1578 	if (!owner || !is_top_waiter)
1579 		return;
1580 
1581 	raw_spin_lock(&owner->pi_lock);
1582 
1583 	rt_mutex_dequeue_pi(owner, waiter);
1584 
1585 	if (rt_mutex_has_waiters(lock))
1586 		rt_mutex_enqueue_pi(owner, rt_mutex_top_waiter(lock));
1587 
1588 	rt_mutex_adjust_prio(lock, owner);
1589 
1590 	/* Store the lock on which owner is blocked or NULL */
1591 	next_lock = task_blocked_on_lock(owner);
1592 
1593 	raw_spin_unlock(&owner->pi_lock);
1594 
1595 	/*
1596 	 * Don't walk the chain, if the owner task is not blocked
1597 	 * itself.
1598 	 */
1599 	if (!next_lock)
1600 		return;
1601 
1602 	/* gets dropped in rt_mutex_adjust_prio_chain()! */
1603 	get_task_struct(owner);
1604 
1605 	raw_spin_unlock_irq(&lock->wait_lock);
1606 
1607 	rt_mutex_adjust_prio_chain(owner, RT_MUTEX_MIN_CHAINWALK, lock,
1608 				   next_lock, NULL, waiter_task);
1609 
1610 	raw_spin_lock_irq(&lock->wait_lock);
1611 }
1612 
1613 /**
1614  * rt_mutex_slowlock_block() - Perform the wait-wake-try-to-take loop
1615  * @lock:		 the rt_mutex to take
1616  * @ww_ctx:		 WW mutex context pointer
1617  * @state:		 the state the task should block in (TASK_INTERRUPTIBLE
1618  *			 or TASK_UNINTERRUPTIBLE)
1619  * @timeout:		 the pre-initialized and started timer, or NULL for none
1620  * @waiter:		 the pre-initialized rt_mutex_waiter
1621  * @wake_q:		 wake_q of tasks to wake when we drop the lock->wait_lock
1622  *
1623  * Must be called with lock->wait_lock held and interrupts disabled
1624  */
1625 static int __sched rt_mutex_slowlock_block(struct rt_mutex_base *lock,
1626 					   struct ww_acquire_ctx *ww_ctx,
1627 					   unsigned int state,
1628 					   struct hrtimer_sleeper *timeout,
1629 					   struct rt_mutex_waiter *waiter,
1630 					   struct wake_q_head *wake_q)
1631 	__releases(&lock->wait_lock) __acquires(&lock->wait_lock)
1632 {
1633 	struct rt_mutex *rtm = container_of(lock, struct rt_mutex, rtmutex);
1634 	struct task_struct *owner;
1635 	int ret = 0;
1636 
1637 	__assume_ctx_lock(&rtm->rtmutex.wait_lock);
1638 
1639 	lockevent_inc(rtmutex_slow_block);
1640 	for (;;) {
1641 		/* Try to acquire the lock: */
1642 		if (try_to_take_rt_mutex(lock, current, waiter)) {
1643 			lockevent_inc(rtmutex_slow_acq3);
1644 			break;
1645 		}
1646 
1647 		if (timeout && !timeout->task) {
1648 			ret = -ETIMEDOUT;
1649 			break;
1650 		}
1651 		if (signal_pending_state(state, current)) {
1652 			ret = -EINTR;
1653 			break;
1654 		}
1655 
1656 		if (build_ww_mutex() && ww_ctx) {
1657 			ret = __ww_mutex_check_kill(rtm, waiter, ww_ctx);
1658 			if (ret)
1659 				break;
1660 		}
1661 
1662 		if (waiter == rt_mutex_top_waiter(lock))
1663 			owner = rt_mutex_owner(lock);
1664 		else
1665 			owner = NULL;
1666 		raw_spin_unlock_irq_wake(&lock->wait_lock, wake_q);
1667 
1668 		if (!owner || !rtmutex_spin_on_owner(lock, waiter, owner)) {
1669 			lockevent_inc(rtmutex_slow_sleep);
1670 			rt_mutex_schedule();
1671 		}
1672 
1673 		raw_spin_lock_irq(&lock->wait_lock);
1674 		set_current_state(state);
1675 	}
1676 
1677 	__set_current_state(TASK_RUNNING);
1678 	return ret;
1679 }
1680 
1681 static void __sched rt_mutex_handle_deadlock(int res, int detect_deadlock,
1682 					     struct rt_mutex_base *lock,
1683 					     struct rt_mutex_waiter *w)
1684 	__must_hold(&lock->wait_lock)
1685 {
1686 	/*
1687 	 * If the result is not -EDEADLOCK or the caller requested
1688 	 * deadlock detection, nothing to do here.
1689 	 */
1690 	if (res != -EDEADLOCK || detect_deadlock)
1691 		return;
1692 
1693 	if (build_ww_mutex() && w->ww_ctx)
1694 		return;
1695 
1696 	raw_spin_unlock_irq(&lock->wait_lock);
1697 
1698 	WARN(1, "rtmutex deadlock detected\n");
1699 
1700 	while (1) {
1701 		set_current_state(TASK_INTERRUPTIBLE);
1702 		rt_mutex_schedule();
1703 	}
1704 }
1705 
1706 /**
1707  * __rt_mutex_slowlock - Locking slowpath invoked with lock::wait_lock held
1708  * @lock:	The rtmutex to block lock
1709  * @ww_ctx:	WW mutex context pointer
1710  * @state:	The task state for sleeping
1711  * @chwalk:	Indicator whether full or partial chainwalk is requested
1712  * @waiter:	Initializer waiter for blocking
1713  * @wake_q:	The wake_q to wake tasks after we release the wait_lock
1714  */
1715 static int __sched __rt_mutex_slowlock(struct rt_mutex_base *lock,
1716 				       struct ww_acquire_ctx *ww_ctx,
1717 				       unsigned int state,
1718 				       enum rtmutex_chainwalk chwalk,
1719 				       struct rt_mutex_waiter *waiter,
1720 				       struct wake_q_head *wake_q)
1721 	__must_hold(&lock->wait_lock)
1722 {
1723 	struct rt_mutex *rtm = container_of(lock, struct rt_mutex, rtmutex);
1724 	struct ww_mutex *ww = ww_container_of(rtm);
1725 	int ret;
1726 
1727 	__assume_ctx_lock(&rtm->rtmutex.wait_lock);
1728 	lockdep_assert_held(&lock->wait_lock);
1729 	lockevent_inc(rtmutex_slowlock);
1730 
1731 	/* Try to acquire the lock again: */
1732 	if (try_to_take_rt_mutex(lock, current, NULL)) {
1733 		if (build_ww_mutex() && ww_ctx) {
1734 			__ww_mutex_check_waiters(rtm, ww_ctx, wake_q);
1735 			ww_mutex_lock_acquired(ww, ww_ctx);
1736 		}
1737 		lockevent_inc(rtmutex_slow_acq1);
1738 		return 0;
1739 	}
1740 
1741 	set_current_state(state);
1742 
1743 	trace_contention_begin(lock, LCB_F_RT);
1744 
1745 	ret = task_blocks_on_rt_mutex(lock, waiter, current, ww_ctx, chwalk, wake_q);
1746 	if (likely(!ret))
1747 		ret = rt_mutex_slowlock_block(lock, ww_ctx, state, NULL, waiter, wake_q);
1748 
1749 	if (likely(!ret)) {
1750 		/* acquired the lock */
1751 		if (build_ww_mutex() && ww_ctx) {
1752 			if (!ww_ctx->is_wait_die)
1753 				__ww_mutex_check_waiters(rtm, ww_ctx, wake_q);
1754 			ww_mutex_lock_acquired(ww, ww_ctx);
1755 		}
1756 		lockevent_inc(rtmutex_slow_acq2);
1757 	} else {
1758 		__set_current_state(TASK_RUNNING);
1759 		remove_waiter(lock, waiter);
1760 		rt_mutex_handle_deadlock(ret, chwalk, lock, waiter);
1761 		lockevent_inc(rtmutex_deadlock);
1762 	}
1763 
1764 	/*
1765 	 * try_to_take_rt_mutex() sets the waiter bit
1766 	 * unconditionally. We might have to fix that up.
1767 	 */
1768 	fixup_rt_mutex_waiters(lock, true);
1769 
1770 	trace_contention_end(lock, ret);
1771 
1772 	return ret;
1773 }
1774 
1775 static inline int __rt_mutex_slowlock_locked(struct rt_mutex_base *lock,
1776 					     struct ww_acquire_ctx *ww_ctx,
1777 					     unsigned int state,
1778 					     struct wake_q_head *wake_q)
1779 	__must_hold(&lock->wait_lock)
1780 {
1781 	struct rt_mutex_waiter waiter;
1782 	int ret;
1783 
1784 	rt_mutex_init_waiter(&waiter);
1785 	waiter.ww_ctx = ww_ctx;
1786 
1787 	ret = __rt_mutex_slowlock(lock, ww_ctx, state, RT_MUTEX_MIN_CHAINWALK,
1788 				  &waiter, wake_q);
1789 
1790 	debug_rt_mutex_free_waiter(&waiter);
1791 	lockevent_cond_inc(rtmutex_slow_wake, !wake_q_empty(wake_q));
1792 	return ret;
1793 }
1794 
1795 /*
1796  * rt_mutex_slowlock - Locking slowpath invoked when fast path fails
1797  * @lock:	The rtmutex to block lock
1798  * @ww_ctx:	WW mutex context pointer
1799  * @state:	The task state for sleeping
1800  */
1801 static int __sched rt_mutex_slowlock(struct rt_mutex_base *lock,
1802 				     struct ww_acquire_ctx *ww_ctx,
1803 				     unsigned int state)
1804 {
1805 	DEFINE_WAKE_Q(wake_q);
1806 	unsigned long flags;
1807 	int ret;
1808 
1809 	/*
1810 	 * Do all pre-schedule work here, before we queue a waiter and invoke
1811 	 * PI -- any such work that trips on rtlock (PREEMPT_RT spinlock) would
1812 	 * otherwise recurse back into task_blocks_on_rt_mutex() through
1813 	 * rtlock_slowlock() and will then enqueue a second waiter for this
1814 	 * same task and things get really confusing real fast.
1815 	 */
1816 	rt_mutex_pre_schedule();
1817 
1818 	/*
1819 	 * Technically we could use raw_spin_[un]lock_irq() here, but this can
1820 	 * be called in early boot if the cmpxchg() fast path is disabled
1821 	 * (debug, no architecture support). In this case we will acquire the
1822 	 * rtmutex with lock->wait_lock held. But we cannot unconditionally
1823 	 * enable interrupts in that early boot case. So we need to use the
1824 	 * irqsave/restore variants.
1825 	 */
1826 	raw_spin_lock_irqsave(&lock->wait_lock, flags);
1827 	ret = __rt_mutex_slowlock_locked(lock, ww_ctx, state, &wake_q);
1828 	raw_spin_unlock_irqrestore_wake(&lock->wait_lock, flags, &wake_q);
1829 	rt_mutex_post_schedule();
1830 
1831 	return ret;
1832 }
1833 
1834 static __always_inline int __rt_mutex_lock(struct rt_mutex_base *lock,
1835 					   unsigned int state)
1836 {
1837 	lockdep_assert(!current->pi_blocked_on);
1838 
1839 	if (likely(rt_mutex_try_acquire(lock)))
1840 		return 0;
1841 
1842 	return rt_mutex_slowlock(lock, NULL, state);
1843 }
1844 #endif /* RT_MUTEX_BUILD_MUTEX */
1845 
1846 #ifdef RT_MUTEX_BUILD_SPINLOCKS
1847 /*
1848  * Functions required for spin/rw_lock substitution on RT kernels
1849  */
1850 
1851 /**
1852  * rtlock_slowlock_locked - Slow path lock acquisition for RT locks
1853  * @lock:	The underlying RT mutex
1854  * @wake_q:	The wake_q to wake tasks after we release the wait_lock
1855  */
1856 static void __sched rtlock_slowlock_locked(struct rt_mutex_base *lock,
1857 					   struct wake_q_head *wake_q)
1858 	__releases(&lock->wait_lock) __acquires(&lock->wait_lock)
1859 {
1860 	struct rt_mutex_waiter waiter;
1861 	struct task_struct *owner;
1862 
1863 	lockdep_assert_held(&lock->wait_lock);
1864 	lockevent_inc(rtlock_slowlock);
1865 
1866 	if (try_to_take_rt_mutex(lock, current, NULL)) {
1867 		lockevent_inc(rtlock_slow_acq1);
1868 		return;
1869 	}
1870 
1871 	rt_mutex_init_rtlock_waiter(&waiter);
1872 
1873 	/* Save current state and set state to TASK_RTLOCK_WAIT */
1874 	current_save_and_set_rtlock_wait_state();
1875 
1876 	trace_contention_begin(lock, LCB_F_RT);
1877 
1878 	task_blocks_on_rt_mutex(lock, &waiter, current, NULL, RT_MUTEX_MIN_CHAINWALK, wake_q);
1879 
1880 	for (;;) {
1881 		/* Try to acquire the lock again */
1882 		if (try_to_take_rt_mutex(lock, current, &waiter)) {
1883 			lockevent_inc(rtlock_slow_acq2);
1884 			break;
1885 		}
1886 
1887 		if (&waiter == rt_mutex_top_waiter(lock))
1888 			owner = rt_mutex_owner(lock);
1889 		else
1890 			owner = NULL;
1891 		raw_spin_unlock_irq_wake(&lock->wait_lock, wake_q);
1892 
1893 		if (!owner || !rtmutex_spin_on_owner(lock, &waiter, owner)) {
1894 			lockevent_inc(rtlock_slow_sleep);
1895 			schedule_rtlock();
1896 		}
1897 
1898 		raw_spin_lock_irq(&lock->wait_lock);
1899 		set_current_state(TASK_RTLOCK_WAIT);
1900 	}
1901 
1902 	/* Restore the task state */
1903 	current_restore_rtlock_saved_state();
1904 
1905 	/*
1906 	 * try_to_take_rt_mutex() sets the waiter bit unconditionally.
1907 	 * We might have to fix that up:
1908 	 */
1909 	fixup_rt_mutex_waiters(lock, true);
1910 	debug_rt_mutex_free_waiter(&waiter);
1911 
1912 	trace_contention_end(lock, 0);
1913 	lockevent_cond_inc(rtlock_slow_wake, !wake_q_empty(wake_q));
1914 }
1915 
1916 static __always_inline void __sched rtlock_slowlock(struct rt_mutex_base *lock)
1917 {
1918 	unsigned long flags;
1919 	DEFINE_WAKE_Q(wake_q);
1920 
1921 	raw_spin_lock_irqsave(&lock->wait_lock, flags);
1922 	rtlock_slowlock_locked(lock, &wake_q);
1923 	raw_spin_unlock_irqrestore_wake(&lock->wait_lock, flags, &wake_q);
1924 }
1925 
1926 #endif /* RT_MUTEX_BUILD_SPINLOCKS */
1927